Interactive van Krevelen diagrams – Advanced visualisation of mass spectrometry data of complex mixtures

The field of complex mixture analysis has advanced significantly in the past two decades, although its history goes much further back. When Dirk Willem van Krevelen developed his now eponymous diagram in 1950 to represent the chemical makeup of coals, he proposed that the chemical nature of samples, including the presence of structural motifs and chemical properties, could be inferred from the elemental ratios of the sample. While his work, limited by the technology of the era, looked at whole samples characterised by the ratio of elements present, i.e. number of carbonsto-hydrogens within the sample, modern mass spectrometry allows us to examine in a similar manner the individual components of a complex mixture. Since 2003,when themodern vanKrevelen diagramwasfirst used to visualise complex MS datasets, every significant high-resolution mass spectrometric analysis of a complex mixture has included one. Today’s van Krevelen diagram places every assigned unique chemical formula on a 2D scatter plot of H/C ratio versus O/C ratio, although other elemental ratios can also be used. Although this represents a break from the original intentions of van Krevelen, themodified technique has become auseful tool for the interpretation andvisualisation of complex data. For example, regions of the van Krevelen plot can be tentatively associated with certain compound classes, such as lipids (O/C < 0.2, H/C 2 – values quoted are approximate), carbohydrates (H/C2,O/C1), or condensed hydrocarbons (O/C < 0.2, H/C < 1). In the field of complex mixture analysis, a number of methods are available to the enterprising chemist; however, Fourier transform ion cyclotron resonance mass spectrometry (FTICR MS) reigns supreme as the ‘gold standard’ technique. Likewise, there exist a number of well-studied complex mixtures, including natural organic matter (NOM), i.e. dissolved organic matter, soil organic matter, and organic aerosols, petroleum, or beverages such as wine or Scotch whisky. Amongst the most complex of these, a component of NOM and the closest sample to a universal standard, is Suwannee River Fulvic Acid (SRFA) produced by the International Humic Substance Society. A typical electrospray ionisation (ESI)-FTICR mass spectrum of SRFA will contain thousands of peaks across a range of masses, predominantly between m/z 200 and 700. Due to its ubiquity and complexity, SRFA was chosen to demonstrate the capability of the visualisation tools described herein. With the mass accuracy of FTICR MS spectra in parts-per-billion, routine and confident assignment of thousands of unique chemical formulae to individual peaks is now increasingly possible. The generation of this volume of data represents a significant challenge in terms of data visualisation, interrogation, and interpretation that has not been addressed so far. Here, we present a handful of tools aimed at filling this gap. We have developed a version of the van Krevelen diagram, which introduces interactivity, and allows the analyst, or reviewer, to interrogate the data in an intuitive way. This interactive van Krevelen, or i-van Krevelen for short, is generated using the Bokeh Python plotting library. The developed tools are fully compatible with data assigned using any software package, as the input for the i-van Krevelen scripts are three text files containing (1) monoisotopic peak assignments, (2) isotopologue peak assignments, (3) remaining unassigned, but detected, peaks. Example input files are included with the suite of presented tools. The Bokeh API allows for the straightforward coding, in Python, of complex JavaScript (JSON) plots as HTML5 Canvas objects. The output from this tool is a standard HTML document compatible with any modern web browser such as Google Chrome, Firefox, or Internet Explorer. The main feature of the i-van Krevelen software is the generation of interactive diagrams including a centroid mass spectrum, van Krevelen, DBE vs carbon number plot and the modified Aromaticity Index vs carbon number plot. The plots are linked together, such that selecting any data points in one plot highlights those same points – i.e. unique chemical formula – in the other plots. In addition, these plots are explorable, featuring zoom and pan tools, as well as a display of the key information of each point in a hover-tool. Finally, the data points can be used as hyperlinks – in our implementation, they link to a ChemSpider (The Royal Society of Chemistry, Cambridge, UK) search for their molecular formula. The benefits of these featureswill be immediately obvious to any analytical chemist who has tried tomake sense of complex static van Krevelen diagrams of complex mixtures. For example, in a standard van Krevelen plot, numerous points may be superimposed if they share elemental ratios but differ in molecular formulae. As a van Krevelen plot is a specific type of scatter plot, it is susceptible to the same problems as other any other scatter plot, and can be misinterpreted when hundreds or thousands of points are plotted. Whilst the addition of colour and transparency can reduce these problems, they are not eliminated entirely. One alternative is to plot data density, not individual data points – i.e. a histogram or kernel density plot in 1D, or a hexagonally binned data plot in 2D. This allows easier visualisation of where the most (or largest, or most intense, depending on the density variable) data points are; however, this approach leads to a loss of information about specific components and their molecular formulae. With interactivity, however, a user can zoom to a region of